You could select the block/matrix of interest "manually" and then pass this preprocessed input to Gnuplot to visualize it. For example:
getMatrix(fName, i) = sprintf("<sed -n -e'%d,%dp' '%s'", 2*i-1, 2*i, fName)
plot getMatrix('test.dat', 2) matrix w image
Here, Gnuplot "sees" only the output of the sed command which filters from the file test.dat only lines 3 and 4, i.e., the matrix you show in your post.
EDIT:
The statement
getMatrix(fName, i) = sprintf("<sed -n -e'%d,%dp' '%s'", 2*i-1, 2*i, fName)
defines a function which for given file name and matrix index (assumed to be 1-based) constructs first a string containing the plotting command which is then passed to plot. For example, for i=1 and file test.dat, the command looks like <sed -n -e'1,2p' 'test.dat'. If one instructs Gnuplot to plot such a "thing", it executes the command after the leading < and plots its output (instead of the entire file test.dat). In this particular case, sed extracts only lines with line numbers (1-based) from 1 to 2 (inclusive).
In order to extend this for matrices of general size N (assuming that the file being processed has corresponding number of columns), one could for example do:
unset key
getMatrix(fName, N, i) = sprintf("<sed -n -e'%d,%dp' '%s'", (i-1)*N + 1, i*N, fName)
do for [i=1:2] {
plot getMatrix('test.dat', 10, i) matrix w image
pause 5
}
The only change here is the extra parameter N denoting the size of the individual matrices and how the range of rows is calculated...