Cliptin, yes, 2M in each direction. The setup at the office in London is a bit odd - they use asymmetric routing for the return path for security reasons. This is done through a pair of 2M lines, each of which is simplex.
As for your question, a T1 (or any other telco line) is fully duplex, of course, so 1.544M is available in each direction. However, the
total bandwidth in the connection is 1.544M as well, so you don't buy a T1 line and magically end up with 1.544M in each direction for a total of 3M.
One thing you may or may not know - available bandwidth on a non-fibre connection (ie. copper, like a T1 or an E1, or subrate stuff) is determined by the clock speed in the CSU/DSU, so the faster the connection rate (64K, 128K, 512K, 2M...), the lower your latency. With fibre the latency is pretty much fixed, but newer cables tend to have lower latencies than older ones (since the speed of light is constant, the improvements have been in the transmission kit).
The company I work for has just lowered the target round trip delays (RTD) for intra USA, trans Atlantic and intra Europe data. This has a lot to do with us running 4 9.6G connections across the Atlantic (we're the first to deploy OC192 between continents) and buying up a bunch of new dark fibre and transmission facilities. A few months ago a tech explained to me that our target RTD SLA for the IP core network is in fact calculated as the speed of light + 30ms to anywhere in the world on the network - just think how fast the core IP switches have to be to keep up with that.
