Chance of  fatigue for limited number of  stimulus (say around 2000) is very less. for example stimulus interval duration of 1000 ms or 1 sec and averaging out 200 waves in BAEP does not show any effects of fatigue.

3-5 sec in BAEP and VEP is fine.

Fatigue really depends on what you're asking the participant to do.  Complex decisions vs. simple judgments vs. passive listening all impact differently.  We've done studies with stimuli intervals of 2-5 seconds without much trouble.  Mismatch Negativity uses very short intervals (350-700ms).  

Stimulus duration really depends on what you're trying to evoke in your paradigm. There have even been perpetual screen displays like the Segalowitz maze paradigm (https://www.brocku.ca/psychology/people/SPR%202005%20posters/Mathewson,%20Dywan,%20Tays,%20Segalowitz%20&%20Snyder%20SPR%202005.pdf).  

I would caution against too many trials as split-half comparisons have long shown significant habituation effects in evoked potentials.  There's a particularly nice figure in David's Regan's book (978-0444013248).  But the literature on habituation, attention, and fatigue in ERPs goes back to the 1950's (at least in my collection of books).  

With an increasing duration of the epoch the risk for (muscle) artificats, especially blinks, increases as well. Depending on your correction methods long trails might this be lost. 

You should have some jitter in the interval timings to avoid the participant thinking that their might be a pattern.  

The delay will largely depend on what your stimuli are and the processes that they invoke.  If you are mixing multiple different stimuli or items into a single trial then beware that whatever comes last should be given enough time for the brain to reset between the tasks.  e.g.  you don't want to have something pop up towards the end that will evoke a late positive potential unless you are taking measures to account for this.