> I recently read a theory that they are using the wrong compiler. This would explain a lot.

It is not just theory but fact that code compiled for OS4 is not tuned for the PA6T via compiler options. If it was, it would either run slower on the other OS4-supported CPUs or you'd need to provide several executables, one for each sub-arch. If you check OS4Depot, you will find that there's only one executable per archive with almost all packages (so either no specific compiles for G2/G3/G4 and PPC4x0, let alone PA6T, or tuned for one of them and thus running extra-slow on the others). So even when disregarding SIMD, compiling for OS4 is mostly a compromise where you target the lowest common denominator, which results in an executable that runs on all CPUs but not as fast as it could on any of them.
The same is true for MorphOS, btw (see the need for G5-specific executables for some applications), and the Sam460 port will only intensify this issue. Amigans know this whole shebang from the m68k days with specific compilation for 68000/20/30/40/60.

"AFAIK the old GCC version 4.2.4 in the current SDK may generate instructions not supported by the PA6T CPU which have to be emulated by an exception handler in the kernel, which is extremely slow."
http://www.amigans.net/modules/xforum/viewtopic.php?post_id=90747#forumpost90747

"there is no support for that processor target in the AmigaOS4 SDK yet."
http://euaejit.blogspot.com/2014/07/ppcjitbeta04b-hundreds-and-xthousands.html

[ Edited by Andreas_Wolf 28.07.2014 - 20:39 ]