Performance engineering is my business. The number of significant factors are as numerous as the stars in the sky... To determine how one method compares to another in any given environment, you have to construct equivalent code sets for each of the tools you are testing, along with appropriate (for your usage) applications, and then run them in a controlled environment. Then you may be able to measure their relative performance.
So, what I am saying is that this is NOT a question that has a simple answer. It is WHOLELY dependent upon your application needs and systems environment. A vs. X may work better for your needs, but X vs. A may be better for mine.
All that aside, Intel writes some very good software, and will usually serve you well, for a price. MPI is well-established and open source, and will also perform well in most scenarios, and cheap at twice the price (free).
FYI, I have spent 30+ years designing and writing large-scale distributed systems code. Personally, I don't use either, but have worked with both.