Pretty straightforward, just use the features in the file to find the wanted sections...it's quite regular in its construction so don't even have to do very much looking for what want--should get you started:
N=cell2mat(textscan(fid,'Number of specimens: %d %*[^\n]','headerlines',11));
while isempty(strfind(l,'_')), l=fgetl(fid);end
Result will be a cell array of N nx3 data arrays. Above at command line here leaves:
> whos d
Name Size Bytes Class Attributes
d 1x3 39564 cell
[532x3 double] [561x3 double] [548x3 double]
Looks about right...
If the sections of auxiliary inputs, etc., are always a fixed length could use line counts in 'headerlines' parameter and avoid the while loops looking for section start, but the above will work for different sizes of those elements as long as the sections remain the same. For files of this size, that won't be that much overhead to process on a record-by-record basis anyway...