The test System.Tests.Perf_Array.ArrayCopy2D(https://github.com/dotnet/performance/blob/0c626c0de72e9e018bb02d0f92b97117098f1fee/src/benchmarks/micro/corefx/System.Runtime/Perf.Array.cs#L236) shows a big slowdown in our array copy compared to .NET core.

• This takes 3284214.702 ns on Mono
• This takes only 342.3348056 ns for .NET core. This is 00.0104% of the time it takes Mono.

A reproduction of this has been made with the executable here: https://gist.github.com/alexanderkyte/3cc2522a0b7391a1326dc65afcd4a3db

When profiled with Instruments, we see a slowdown due to some handle manipulation. It doesn't seem to be falling back to any intrinsic.