If you use the stimulus files indexed as 0's and 1's, I think that the best number of max lag is usually not the biggest one as multicollinearity allows. Instead, people usually pick the one they think reflects the real delay of hemodynamic response. Of course in reality you would try a few numbers to get the best fitting and statistics.
Yes, with 4 concatenated runs and a max lag of 4, 000000001111100000 (not 0000000011111) seems to be the correct vector for glt to get the contrast between the first condition (or task) vesus baseline.
With wavered input for 3dDeconvolve, you assume the impulse response function is gamma variate, and with gamma variate the only flexibility is that you can adjust a couple of parameters (including response delay) when you run waver. With stimulus files indexed in 0's and 1's, the program 3dDeonvolve will fit your data with various impulse response functions and you can control the fitting goodness by setting the minimum and maximum lags.
Gang
