It's just a guess, but I think the biggest limitation to the experimental area is the "face validity". In animals, it is probably difficult to generalize from any model; In humans, little could be qualitatively analyzed.
I agree, the small sample sizes, in older people the differential diagnosis delirium versus dementia, the difficult determination of acute onset and fluctuating course, and, when you are trying to establish an etiologic factory, the multifactorial genesis.
The main problems in running a study are 1. low prevalence 2. difficulties with consent & ethics approval 3. difficulties with recognition/recruitment 4. difficulties with case definition. The main problems in interpreting delirium studies are 1. difficulties with case definition. 2 low sample size 3. bias in recruitment towards either milder cases (who consent) or severe cases (who are recognized) 4. failure to allow for variations with time. Hope that helps!
Regarding ICU delirium, the most common limitations are the different evaluation tools used, because some are more useful in detecting subsyndromal delirium than others, also the sedation regime used in the different ICU´s, because if there is a high prevalence of deep sedation then the evaluation is not feasible for a large number of days... Then, there is a problem when it is time to translate evidence to daily practise, because as there are so many possible confounders, there are lots of conditions usually excluded (for example, dementia) from studies, but we still have such patients in our ICU... Hope it is useful!