A .bxh (or XCEDE) file is only a wrapper to data in other files. So the
wrapper in the DICOM case is going to be a lot larger because in the GE
case it is pointing to many (sometimes thousands) of DICOM files.

In the pfile2bxh case, it only parses the P-file header for the image
metadata, not the actual image data. If you want to give it image data
to wrap too, you need to specify it as a separate input file:

pfile2bxh PFILE IMAGEDATA.raw WRAPPER.bxh

The P-file generally contains k-space data, not reconstructed image
data. For our in-house pulse sequences, we recon the k-space data from
the P-file into a raw image data file and then use the above command to
get them all wrapped together by the same .bxh file. For sequences that
generate DICOM, we just use the DICOM (though there are a few fields in
the Pfile header that we wish were in the DICOM, so we might soon take
advantage of both and merge them).

If I am misunderstanding your problem, please let me know!