Wow. Read 'datacenter' for 'machine in the local network'.

Yes; more accurately, read "anywhere in EC2 to probably anywhere else in EC2".

Just so everyone knows, this specific attack is rather old. Modern versions of the RSA algorithm use blinding to specifically prevent this, which causes the RSA calculations to effectively be performed on random data.

There has also recently been a published timing attack regarding elliptic curve signatures, and there are some older papers about AES's vulnerability to timing attacks due to cache misses when using S-boxes. So while this paper specifically isn't a viable attack today, it is still relevant.

This was published in 2003: http://portal.acm.org/citation.cfm?id=1251354

That doesn't change the importance of securing against timing side-channel attacks, however.

In fact, the attacks have gotten much better since then; for instance, see Crosby & Wallach's paper on using statistical filters to establish nanosecond timing bounds on "local" (ie, cross-datacenter) networks:

http://www.cs.rice.edu/~dwallach/pub/crosby-timing2009.pdf

In other words, don't use versions of OpenSSL that are almost a decade old.

It is a novel class of attacks, and bites many people squarely on the hindquarters. For example, every available Java OpenID implementation was timable back in 2010.