Only read the first and last page if the paper, but here was my quick takeaway.
There are 2 things the attacks do: get the victim data into cache, then read it out. This paper just shows a different way to read it out. Instead of flushing the cache it induces false sharing to ping pong the cache line so instead of waiting for main memory to service the request you are waiting on the cache coherence protocol to catch up.
These were already described attacks on the web, so i don't consider this new although they aren't talked about much probably because they are a little harder to understand.
Since the fixes to both spectre and Meltdown are focused on the first part, stopping data getting into the cache, they are work for this new read method.
Meltdown: fix the hardware permission bug on cache hit (probably very easy) and everything will be fine.
Spectre: just as out of order execution can cause funny things in threaded programs, they can in non threaded programs too it seems. Devs just need to be aware of it and guard against it. Spectre is basically a process reading its own memory, and we shouldn't be surprised by that.