The math program MLAB is short on Mac amenities but is quite powerful for a large range of science applications. This includes any activities in which the expected behavior model of the underlying physical, chemical, or biological system is an ordinary differential equation or system of equations. That's a big range. In fact, MLab is more than a match for Prescience's Theorist, a middleweight symbolic-math package; moreover, in its own domain MLab shows several advantages over Wolfram Research's Mathematica.
MLab, which was developed at the National Institutes of Health, is not a general-purpose math program. It's essentially a programming language, drawing on constructs from Pascal, FORTRAN, and BASIC, that implements a large assortment of commands for fitting experimental data to dynamical (usually differential-equation) models. MLab also covers statistics (descriptive, cluster-analysis, and survival analysis), fast-matrix algebra, and an assortment of graphs oriented toward presentation in scientific journals.
By Mac software-interface standards, MLab is pretty primitive---its Unix and DOS origins make it a good choice for researchers familiar with text-based command-line programming. At start-up MLab displays a blank console window with no horizontal-scrolling or sizing capabilities. The File menu doesn't even include old favorite commands such as Open. The program requires a 68020 or higher processor and hardware math coprocessor (the on-CPU hardware math in the Quadra 650, 800, and 950 works), and it doesn't yet come in a Power Mac version. However, it can function in less than 2MB of free RAM, making it the least-demanding high-powered science application for the Mac.
The payoff in MLab, compared with other scientific software, lies in the large number of carefully explained, easy-to-program examples included in its truly superior documentation. The tutorials on enzyme kinetics, multicompartmental pharmacological models, pulsed magnetic-resonance spectroscopy, nonlinear oscillators, and nearly 30 other scientific topics are clear and full of practical examples. In principle, you could do this sort of modeling in Mathematica or many other packages with well-designed floating-point numerics. The difference is that MLab tells you explicitly how to do it, with instructions written by scientists for scientists.
The Last Word MLab lets you do fast computation on numerically complex real-world problems, using only an ancient Mac II or higher. You must be willing to write short (usually single-page) programs, but MLab will reward the professional researcher with its mathematical-modeling capabilities.