How big is 'huge'? Assuming I've understood your requirements, your accepted solution seems terribly inefficient, memory-wise, and doesn't scale particularly well.
An alternative approach might be to interpret each r-consecutive digits as a binary integer representation, use a sliding filter for a binary to decimal conversion, then check to see if any integers in one set match the integers in the other.
I think the following captures the edge cases correctly, but you might want to test it further.
A = rand(1.0e4, 1) > 0.5; A = num2str(A); A(A == ' ') = ;
B = rand(1.2e4, 1) > 0.5; B = num2str(B); B(B == ' ') = ;
r = 6;
matches = bsxfun(@eq,A,B.');
intv = (0:r-1)*(size(matches,1)+1)+1;
idx = find(matches);
idx = idx(idx <= max(idx) - max(intv));
out = any(all(matches(bsxfun(@plus,idx,intv)),2));
bin = 2.^(0:r - 1);
A2 = filter(bin, 1, A == '1');
B2 = filter(bin, 1, B == '1');
bool = any(ismember(A2(r:end), B2(r:end)));
out == bool
output (note that I wrapped these in a function to ensure JIT acceleration in version 2014a):
Elapsed time is 5.936476 seconds.
Elapsed time is 0.002765 seconds.
For sizes in the range of 1.0e5, my computer runs out of RAM using the accepted method and starts page swapping, which is impressive since I have 256GB of RAM (I got tired of waiting for a result); for the proposed solution, it only takes 0.022 seconds.
Note that if numel(A) is small (e.g. < 30), the accepted solution may indeed be faster.
The second approach could probably be made faster still, however, if you weren't starting from strings (e.g. if instead you used logical arrays for A and B).
Let me know if you see an error in the logic of the proposed solution, or if it doesn't work for your use case. (And if you find it useful, you could still upvote it for posterity.)