Ick. Currently only the header is byteswapped if the machine stamp
says the endianness is wrong. It seems like the data probably should
be too. And then the heuristic applied.
MRC files suck, so whatever you do will only work some of the time. You
will also find MRC files with byte-swapped headers but not byte-swapped
data, for example. There are also two different file types that use
'MRC' format (some of the header fields mean entirely different things)
but of course there is no way to reliably tell the difference. The best
we can do is to write out correct files, and do our best to figure out
the inputs. If you're going to mess with the heuristics, Frido's
suggestion is the best one - you add the suspect MRC input file to the
testcases, so that further tweaks to the heuristics don't break existing
files. I believe EMBED used my testcases for Modeller's MRC reader, but
these can certainly be expanded.