It all depends how often you solve linear problems. If you are occasional user any package for any language will do.
If, however, you process tons of sparse, dense, categorical, or continuous data BE WARY: "lapack", "eispack", "linpack", "scalapack", and "lis" will give you different solutions. Occasionally, crash on some machines. Compilers, linkers, precision, and missing values matter here, too.
linpack and eispack are vestigial. They were folded into lapack, which is the "one true package" now (e.g., http://en.wikipedia.org/wiki/EISPACK). The situation here is really not complicated.
scalapack is a different animal entirely. It's a parallel version of lapack (via MPI) so should not be expected to duplicate lapack results because partitioned algorithms will be used. But if you're using scalapack, you know that already.
If, however, you process tons of sparse, dense, categorical, or continuous data BE WARY: "lapack", "eispack", "linpack", "scalapack", and "lis" will give you different solutions. Occasionally, crash on some machines. Compilers, linkers, precision, and missing values matter here, too.