My $0.02 on RMAA:
The program is very good, albeit a bit limited on what it does (i.e., what you get is what it gives you, with very few customizations). The real variable, though, is the sound card you use with it. If you don't use with a very good sound card then you're basically measuring just the sound card rather than the amp in question.
Doing a loopback test on the sound card alone is telltale. Most built-in sound interfaces on computer motherboards are very poor candidates for use with RMAA, because they perform so badly. Even some "pro"-grade cards are inadequate. Cards that are good for gaming, etc., are not necessarily good enough for RMAA use, because the emphasis is on certain features rather than absolutely the flattest frequency response, lowest noise floor/THD/IMD, and greatest stereo separation across the entire audio band.
I recently switched from using an M-Audio Transit USB to a M-Audio Firewire Audiophile. Both cards are excellent, but the Transit only allows measuring at 24b/48KHz max (and it has a bit more distortion at that mode than at 16b/44.1K), while the FW Audiophile can go full tilt at 32b/96K. The FW Audiophile outperforms the Transit for RMAA purposes by a good margin especially for noise floor and stereo crosstalk due to the higher available bit-resolution. The higher sampling frequency also gives better high frequency response. Being external "mobile" interfaces, the sensitive audio circuitry in each of these can also be situated far away from the noisy internals of the computer.
Another thing about using RMAA is to make sure the measurement environment is wholesome. This means using short and well shielded interconnect cables, turning off fluorescent lights or lights with pulsive dimmers, orienting the setup away from CRT monitors, transformers, motors, or other emissive devices, etc. Those could all affect the measurements in some manner.
I have M-Audio Transit and FW Audiophile sound card loopback test results here if you're interested in seeing them:
http://www.amb.org/rmaa/
For headphone amps, it's important to run the tests with various dummy loads connected to the amp's output. Some amps behave well with high-Z loads (or no load), but the performance deteriorate badly when loaded down with low-Z (i.e., 32 ohm) loads.
RMAA is a useful tool, but should not be the
only tool for measuring the performance of an amplifier. Some features are sorely lacking. For example, it would be nice if RMAA could plot THD vs. frequency and THD vs. output level graphs. Also, RMAA does not (and cannot) measure an amp's pulsive response such as square waves, slew rate, rise time, etc. For some of these there is no substitute for a good function generator, oscilloscope, and some dedicated instruments.
Since the sound card and test environment could affect the results, any published RMAA measurements of amplifiers should always be accompanied with the sound card loopback results for comparison (in order to establish a frame of reference). Since the same amp tested via two different setups can yield different outcomes, RMAA results should only be used to check for obvious flaws, not to compare for absolute specs. For example, if one amp has 0.001% THD tested on one setup, and another amp got 0.003% on a different setup, it's not meaningful to conclude whether one actually has lower distortion than the other.