I have a couple of remarks which should hopefully bring the issues you point out into focus. First, the Maple kernel is actually very small. It implements only a few mathematical algorithms (such as normal). For the most part, it's job is to implement data structures: sums, products, hash tables, rtables, etc, and the Maple programming language: procedures, branching, loops, as well as other primitives such as integer and finite field arithmetic. For a reference I suggest the book "Algorithms for Computer Algebra" by Geddes, Czapor, and Labahn, as well as the Maple programming guides distributed with the full version of Maple. You can download the programming guides here. These are all good introductions to the data structures implemented in the Maple kernel.

About 90% of the mathematical functionality of Maple is written in the Maple language, making the system extremely flexible. You can build almost anything on top of the Maple kernel, and people often do. There are programs and packages for roughly every area of mathematics, including some very specialized areas, and routines for engineering, the life sciences, and education as well. These are all written in the Maple language, and they are bolted on to the system, often independently of oneanother. This includes things like mathematical functions and types (!). The actual development of Maple is spread across labs all over the world, for example, in Moscow, Paris, Vancouver, London Ontario, and of course Waterloo. These labs contribute most of the specialized packages and routines. The Maplesoft company maintains and enhances the core routines (such as simplify, int, solve, etc), develops the Maple kernel, and oversees development of the system as a whole.

The picture I'm trying to paint here is that Maple is more like an organic ecosystem than a monolithic product. The company has built a platform and it supports many different (and sometimes competing) development efforts. The advantage of this approach is that the good routines win out over the long term, and the product is allowed to evolve and adapt to new uses quickly. This is the reason why Maple is so widely installed - there is almost always something useful that it does. The downside of this approach is that it is chaotic. Old parts of the system don't seem to know about new functionality, and some newer parts occasionally conflict. Long-term users of Maple accept this, and complain to the company endlessly about what needs to be fixed. The important thing is that the company actually does fix things, so that over time some coherence begins to emerge. The added benefit is that the product is developed in a direction that is influenced strongly by users.

As for your specific complaint, it looks like you found a rough spot in the system. The old linalg package (which uses vectors and lower-case matrix) was replaced by the LinearAlgebra and VectorCalculus packages (which use Vector and Matrix). There is an incompatibility there because the VectorCalculus package supports a more general notion of Vector (including, for example, cylindrical and spherical coordinates) that don't make sense outside of that package. Where possible, you should try to stay within the VectorCalculus package if you are using those kinds of features. For example, instead of a list try to use a Vector of functions. For functions which do not understand Vectors (such as int), use the map command to apply the function to each element. For something more complicated (such as surface integrals) you will have to write your own procedures if it is not built in to the system. Usually these are not too difficult to write, but post details (or better yet, an example calculation) if you need suggestions.