Currently sequential discovery of tests with MaxCpuCount = 1 is faster than parallel discovery (MaxCpuCount = 0) for test assemblies that only contain a few (100) tests. This is because -

1. Sequential discovery reuses the same testhost process to discover all supplied test sources. Parallel discovery on the other hand creates a new testhost process for every supplied test source. The overhead of process startup, extension loading etc. is significantly greater than the time required to discover a few (100) tests in this case. Sharing / reusing each teshost processes to discover more than one test source over the lifetime of the discovery operation should improve performance significantly.

2. The testshost processes for parallel discovery are started on a staggered basis and the actual number of testhosts running at any point is always (significantly) smaller than the specified MaxCpuCount. Starting the max allowed number of testhost processes up front should help to better amortize the startup time overhead for individual processes.