MPPENC used to perform well in EAQUAL tests, reported quality increased with a higher NMT etc. That was up to 1.01j: 1.02 performs quite bad.
Reported bandwidth by EAQUAL also dropped a lot on my test clips (from ~21K to ~12K at any quality setting after a few re-encodes).
Also comparing 1.01j to 1.02 (and now 1.04) the newer versions produce a lot more (and more severe) decoding overloads.
Repeated re-encoding make the above problems much more obvious.

Now, the EAQUAL results might be the result of better psychoacoustic modeling in the newer versions of MPPENC, but decoding overloads are really annoying, imposing a higher attenuation to avoid clippings.