After producing models from a quite large amount of training data, I plugged the results into sphinx3 decode and was not pleased to see the following results:

INFO: cont_mgau.c(88): Reading mixture gaussian file '/users/kevin/sphinx3-0.1/model/am/means'
INFO: cont_mgau.c(211): 1 mixture Gaussians, 256 components, veclen 39
INFO: cont_mgau.c(88): Reading mixture gaussian file '/users/kevin/sphinx3-0.1/model/am/variances'
INFO: cont_mgau.c(211): 1 mixture Gaussians, 256 components, veclen 39
INFO: cont_mgau.c(232): Reading mixture weights file '/users/kevin/sphinx3-0.1/model/am/mixture_weights'
FATAL_ERROR: "cont_mgau.c", line 270: /users/kevin/sphinx3-0.1/model/am/mixture_weights: #Mixture Gaussians(6225) doesn't match mean/var parameters(1)

It appears that the parameter files SphinxTrain produced aren't even consistent among themselves!  What went wrong?