Those mu and sigma values don't look right for lognormal distributions. These distributions have only positive values, so they have to be really wildly skewed to have such small means with such large standard deviations. Also, the cutoff of 5 looks too small for these parameters, because a and b will virtually never give a result satisfying the condition. But I think that something like this would work if you adjust the parameters:
mu_a = 0.06; sigma_a = 0.50;
mu_b = 0.08; sigma_b = 0.35;
cutoff = 1020;
nWanted = 10000;
nBatch = 1000;
storedA = [];
storedB = [];
while length(storedA) < nWanted
a = lognrnd(mu_a,sigma_a,nBatch,1);
b = lognrnd(mu_b,sigma_b,nBatch,1);
y = exp((b-a)./a)*1000;
keep = y < cutoff;
storedA = [storedA; a(keep)];
storedB = [storedB; b(keep)];
mean(keep)
end
storedA = storedA(1:nWanted);
storedB = storedB(1:nWanted);
