Generally Quantum mechanics divides a system into what is to be observed and an observer.
This is usually taken to be some macroscopic measuring device (sometimes taken to be some human being in which a special role is assigned to consciousness). But why restrict it to a macroscopic entity? Why not simply another particle?
Now, is there a good physical reason or philosophical reason for this to be dismissed as not sustainable?
All you need is a system that is not in thermal equilibrium (which is practically any macroscopic system), or equivalently a system into which you can inject a reasonable amount of entropy. This process is termed quantum decoherence. The mathematics is nontrivial, but the bottom line is that it's essentially a statistical property, and as such a single other particle is typically insufficient while lots of particles are typically sufficient.
I'm not sure how to explain it in more detail than this without an assumption of way more knowledge of physics. (The Wikipedia article does a decent job (as usual), if you have adequate background.)
The concept of "an observer" does not appear within QM, but on its borders. In fact, it is better thought of not as the edge of QM, but rather the edge of classical mechanics. It is the concept of how classical mechanics can observe behaviors which are too coherent to otherwise explain with classical means.
The Observer Effect is not well defined, because it hasn't been worth anyone's time to define the exact behavior, but it is generally accepted as a pattern where the interaction with non-coherent particles from outside the system cause the waveform to "collapse" to a stable value which is classically measurable.
By this definition, one could define entanglement as being remarkably close to the observer effect. You take two particles which are decoherent, and act on them to cause them to cohere. You could argue that one particle ends up "observing" the state of the other.
However, I find most interpretations of the observer effect rely on the Central Limit Theorem. Since the distributions are finitary both in mean and variance, repeated bombardment of a system with "observer" particles will eventually generate a normal distribution curve which can be measured. However, the rules of statistics break down around N=1, so whether you consider a single particle to be an observer is really a matter of personal judgement.
Before I can answer a brief digression is necessary.
Start of digression.
Some particular "interpretations" of quantum mechanics make the division you describe, but the division is neither necessary nor desirable. Different interpretations of quantum mechanics tend to fall into two categories. Some simply deny the theory's implications (e.g. - Copenhagen), others say that it is false and call for its replacement (e.g. - GRW). The former ought to be rejected out of hand. If the implications of quantum theory disagree with reality then we ought to face that squarely and reject the theory, if it is true then it is important that we should understand it to be able to use it to predict and understand stuff. Theories that disagree with quantum mechanics ought not to be called interpretations of it for clarity.
When QM is applied to macroscopic objects, quantum theory implies that multiple versions of those objects exist that can't interfere with one another as a result of decoherence. Those versions are sorted into layers that are approximately autonomous from one another and approximately obey the laws of classical physics. These layers are called universes and the whole of physical reality is called the multiverse. The multiverse has structure in addition to the universes, so it is not accurate to describe the multiverse as just a collection of parallel universes:
http://arxiv.org/abs/quant-ph/0104033.
End of digression.
You ask whether a measuring instrument can just be another particle. If two systems interact with one another and not with anything else they will be entangled and some distinctively quantum mechanical effects can occur. You can make them interact in such a way that each individual particle will be able to undergo quantum interference. And they can exhibit Bell-type correlations of a sort that are much celebrated, but usually very badly misunderstood:
http://arxiv.org/abs/quant-ph/9906007.
Interference involves a system starting out with one of its observables sharp: that observable has the same value in all of the versions of the particle involved in the experiment. The observable becomes unsharp, undergoes some motion and becomes sharp again in a way that depends on the motion it underwent while it was unsharp. If you try to make a record of that observable's value while the particle is unsharp, then you prevent the interference. For more details see
http://arxiv.org/abs/quant-ph/0703160.
A measurement should be described as an interaction that makes a record that will be preserved indefinitely since that makes it easy to use the measurement result for experimental tests. Since you can undo the interaction between the two particles and erase the record, that doesn't seem to fit the bill. And the existence of permanent records, not some alleged classical level in physics explains why you don't see quantum mechanical effects with objects like coffee mugs.
So the reason to not regard an interaction with another particle as a measurement is a mixture of physics and methodological limitations on what should count as a record.
My two cents:
I just read an article about quantum erasor. In principle what they showed there, is that the quantum interference in a double split experiment vanishes if you force the particle to show its particle property. (This is done by putting the observation screen pretty close to the double split). With a more complicated setup they argue that quantum interference patterns disappear because you take information (which-way-information) out of the system. They proved that by showing that the information, carried by photons, may be erased before it leaves the whole setup. The buzz word for that is delayed choice quatum erasers (a link to Wikipedia which I haven't yet read).
By that I would conclude that particles are (intermediate) observers, because they carry information.
