TACO runs on several platforms, namely the multiple threads template library (MTTL), directly on the PM2 user-level communication layer of SCore, and MPI. The platform can be chosen with the compile time switches -DTACO_MTTL, -DTACO_PM and -DTACO_MPI.
In the case you have SCore 3.XXX or 4.XXX installed, you should use the default version based on the MTTL. It is most efficient as well as robust and offers the best integration into the SCore system.
Otherwise, if you just want to test TACO and its programming model, without installing the rest of the software you might want to try the MPI-based version. It should be portable and is reasonably efficient, too (depending on your MPI implementation). However, on our clusters it is no real match for the MTTL-based implementation.
The third platform available is directly based on SCore and its PM2 communication layer. It offers a similar performance like the MTTL based implementation, but is currently not as well integrated into the score environment. So far, this platform is basically used to study and tune various aspects of TACO in conjunction with user-level networks. You might try this one, too. Sometimes it yields better performance as the MTTL based version sometimes not. Give it a try.
In the file include/taco/TacoConf.h you'll find several tuning parameters as well as their description. Each parameter can be overridden individually. Please have a look at that file and the documentation provided in the comments.