I believe these claims are based on sunspider, which is not really considered a great benchmarking suite. Something like Mozilla's Kraken or Google's Octane may give better real-world results.

FWIW, I only recently understood what made a benchmark suite 'bad'. Basically VMs often speed up by identifying patterns which can be optimized. But there's a balance because this analysis itself has a cost and if the speed up is not much (or does not happen often), then it can slow down other apps. Apparently SunSpider has a number of places where it's easy to add special optimizations, but the result may actually degrade more common code execution.

All browsers optimize for SunSpider, but relying on those numbers alone can be deceiving.

Yea the fact that they are still making update option is going to keep fragmentation an issue for IE, at least, from what I understand, with IE11+ they are making it an opt-out not opt-in