Here's an interesting PDF to read: How speedy is SPDY?.
I'll spoil the results for you, but still encourage you to read the entire PDF to see how the conclusions were built.
-- We experimented with SPDY page loads over a large parameter space
-- Most performance impact of SPDY over HTTP comes from its single TCP connection
-- Browser computation and dependencies in real pages reduce the impact of SPDY
-- To improve further, we need to restructure the page load process
The last point, about having the restructure the page load, fits in nicely with my article on architecting websites for the HTTP/2 era, because it will require restructuring.
The end-result can be anything from a 10% to 80% improvement, in terms of perceived latency and bandwidth consumed.
While this test was performed on SPDY, and not HTTP/2 directly, I believe the end-results can be similar. But to be certain, we'll have to benchmark HTTP/2 directly instead of SPDY.