if you look closely at the output, you will see that the "dots" aren't individual glyphs, but small drawn line segments, usually square. i haven't examined the code that generates them, but i have determined that the size of the last line segment seems to depend on the width of the widest element in the matrix. for example, if you replace B
with M
, the last "dot" is the same size as all the rest. (more experimentation or investigation is needed to see if there's a lower or upper boundary on the effective width.)
with the knowledge that the effect is width-dependent, it's possible to adjust the effective width to get a nice result. adding some tiny spaces or decreasing the space by a tiny bit around the elements yields a more pleasing result:

\documentclass[12pt]{article}
\usepackage{easybmat}
\begin{document}
\Large
$\left[
\begin{BMAT}(@){c}{c.c}
A \\ B
\end{BMAT}\right]$
%
$\left[
\begin{BMAT}(@){c}{c.c}
A \\ \mkern1mu B\mkern1mu
\end{BMAT}\right]$
%
$\left[
\begin{BMAT}(@){c}{c.c}
A \\ \mkern-1mu B\mkern-1mu
\end{BMAT}\right]$
\end{document}
note that the adjustments are size-dependent (hence the use here of the article
class with [12pt]
and \Large
to get a clear image), so some experimentation may be needed.
i will try to figure out where the non-square line segments are being drawn in the package, but this won't happen immediately. the code is definitely flawed in this respect. (and the package is quite old, so an update may not be likely. there may also be a better/newer approach that i don't know about.)