Using data.table in R and taking the DT[J("a")] example at face value I ended up with a really slow function.

faster_use_of_data_table.r

duplicates.step2$days.between  <- sapply(
	1:nrow(duplicates.step2),
	function(q) {
		step2id <- as.character(duplicates.step2.dt[,id][q]) #as.character is important bit!
		temp  <- min(
			difftime(
				duplicates.step2.dt[, Date][q],
				step1.dt[step2id]$Date, #No need to crazily create another data.table
				units=c("days")
			)
		)
	}
)

slower_use_of_data_table.r

step1.dt  <- data.table(step1)
setkey(step1, id)
duplicates.step2$days.between  <- sapply(
	1:nrow(duplicates.step2), #For each row in duplicates.step2
	function(q) {
		step2id <- duplicates.step2.dt[,id][q] #Gets the id
		temp  <- min(
			difftime(
				duplicates.step2.dt[, Date][q], #Gets the date
				step1.dt[J(eval(quote(step2id))]$Date, #gets the dates for all step1 rows of the same id. This is where it all goes wrong
				units=c("days")
			)
		)
	}
)