I benchmarked (using benchmarkdotnet) some of the common sorting algorithms in C# .NET Core 3.0. If you use generic arguments + structs to pass in the comparison function, you can achieve faster custom sorting than than the provided Array.Sort<T>(T[], Comparison<T>) or Array.Sort<T>(T[], IComparer<T>) methods in the .NET libraries. I didn't even replicate the sorting algorithm, but the base Quick and Merge sort algorithms were already faster than the aforementioned .NET methods for at least N ~ 10000 when you use structs (I only benchmarked up to 10000 so far). The source code is available if you want to run the benchmark yourself.

• The XXXRunTime benchmarks use runtime delegates for their compare functions.
• The XXXCompileTime benchmarks use structs for their compare functions (so there is no delegate at runtime).
• The SystemArraySort benchmark is the System.Array.Sort method. Note: the System.Array.Sort method does not allow custom sorting while all other benchmarked methods do.
• The SystemArraySortDelegate benchmark is the System.Array.Sort<T>(T[], System.Comparison<T>) method, where System.Comparison<T> is a runtime delegate.
• The SystemArraySortIComparer benchmark is the System.Array.Sort<T>(T[], System.Collections.Generic.IComparer<T>) method.

Source Code: https://github.com/ZacharyPatten/Towel/blob/master/Tools/Towel_Benchmarking/Sort.cs

BenchmarkDotNet=v0.11.5, OS=Windows 10.0.18362
Intel Core i7-4790K CPU 4.00GHz (Haswell), 1 CPU, 8 logical and 4 physical cores
.NET Core SDK=3.0.100
  [Host]     : .NET Core 3.0.0 (CoreCLR 4.700.19.46205, CoreFX 4.700.19.46214), 64bit RyuJIT
  Job-BDOEBU : .NET Core 3.0.0 (CoreCLR 4.700.19.46205, CoreFX 4.700.19.46214), 64bit RyuJIT

InvocationCount=1  UnrollFactor=1