It was pointed out the when we allocate a check we first optionally poison it, and then decommit all pages following the header.

This may not be optimal, for two reasons:

• we don't need to poison memory if it's going to be decommitted
• it may be preferable to keep one (or more) pages around if we're going to allocate into them immediately

Note in the GC 'decommit' doesn't mean a hard decommit, it tells the OS that it may reuse any physical memory without decommitting the address space.

Note also the currently decommit may not be enabled depending on the platform (see bug 1850741).