I think the problem did not manifest as a bug, since the only
consequence was to generate extra code to return heap number
results in case of the INT32-stub.
Review URL: http://codereview.chromium.org/7192006

git-svn-id: http://v8.googlecode.com/svn/branches/bleeding_edge@8321 ce2b1a6d-e550-0410-aec6-3dcde31c8c00