We can observe a regression on MacOSX x64 for the benchmark Crypto-RSA.sign in the magnitude of 2%.

This change does not appear to do anything that can directly influence the performance in Crypto-RSA.sign so it probably affected alignment in some indirect way. We discussed internally in the performance team whether to file this or not, and agreed it should be reported to see if one can force padding/alignment in a way to avoid this regression from appearing again. There are tools in `src/hotspot/share/memory/padded.hpp` for that purpose.