Because 455^100 is a computation done in DOUBLE PRECISION. Assuch, it goes well beyond the limits of what a double can represent exactly.
But we can do it using extended computations, using the symbolic toolbox, or my own VPI toolbox.
MATLAB computations using a double are NOT carried out to full precision. If they were, you would wait forever to do some simple computations. So a double can represent integers as large as 2^53-1 exactly. (Actually, it gets 2^53 correct also.) But beyond that point, that number will not be represented correctly, exactly, as an integer. And of course 455^100 is well beyond that point, almost to the point where it will overflow into an inf.
The symbolic toolbox has a powermod function in it though, if you have that TB. (VPI also supplies a powermod utility.)