The slowness is caused by the fact to you expand the matrix s and sometimes the matrix data2 in each iteration. You should preallocated these. I also removed the if-end construction as you can easily weed out the unwanted values of data2 afterwards.
x=data(1:end,1); y=data(1:end,2); z=data(1:end,3);
x1=5; x2=9; y1=10; y2=14; z1=25;
mainslope=(y2-y1)/(x2-x1);
count=length(x);
% pre-allocation
s = nan(count,3) ;
data2 = nan(count+1,3) ; % apparently, data2 can become 1 longer than s ...
data2(1,:) = [x1 y1 z1];
for i=1:count
x3=x(i); y3=y(i);
slope=(y3-y1)/(x3-x1);
s(i,:) = [x3 y3 slope] ;
data2(i+1,:) = [x3 y3 z(i)] ;
end
TF = s(:,3) == mainslope ; % OR abs(s(:,3)-mainslope) < 0.001 ???
data2 = data2([true ; TF],:) ; % only retain when slope equals mainslope (plus the first row)
