I'm not sure other software is able to do it either... maybe something that generally exerts more control over the design itself, like PADS Router (and, probably, most offerings in the order-of-magnitude-higher-cost tier... but, yeah..).
Which kind of hints that, like I said... if you can have the autorouter do it... otherwise, yeah...
If nothing else, you can turn off blocking by pressing SHIFT+R while routing (cycles route mode, you want push, or ignore obstacles).
If you're getting rounding errors (yeah, I know the feeling...
), maybe increase it marginally while routing, then drop it afterwards. Silly hack, and doesn't really help once routed, but..
Anyways, only useful comment I really have: do you really need to squeeze the traces? If it's about impedance, remember, differential impedance is, for the most part, a silly idea. Edge-coupled traces have low coupling, and are better thought of as two independent traces being routed through the same path length and surrounding environment (so that noise sources induce equally, and cancel out differentially).
Which is also a reminder not to leave the common mode floating, but to terminate it if possible, and to avoid gross CM noise sources (like, running over a split plane with the potential for large RF voltages across the gap!).
Most e.g. LVDS transmitters have a modest CM source impedance, around 500 ohms, so the transmission line is weakly terminated at either end, so will magnify CM noise by a low Q factor. This is much better than true CCS drivers would manage, so it's not terrible. ECL output impedance is probably on the low side (and tends to have poor CMRR unless used differentially), so would probably benefit from termination.
Tim