Per @kbroniek suggestion

OK. It will be nice to add a benchmark, but I know it will consume your time that we don't have these days.
Thought you could implement this as .net events or take advantage of source code generation. It could be fast then.
What if we add multiple handlers in the multi-threading app. Consider using ConcurrentDictionary.

Do benchmarks and optimise the code. Consider making pipelines more generic to also apply them to command and query handling.