Alright, revised it!

SELECT

ID,

PARSENAME(REPLACE([Subject],';','.'),1) AS Split1,

PARSENAME(REPLACE([Subject],';','.'),2) AS Split2,

PARSENAME(REPLACE([Subject],';','.'),3) AS Split3

INTO #SplitTable

FROM YourTable

SELECT ID,

CASE WHEN Split2 IS NULL AND Split3 IS NULL THEN Split1

WHEN Split3 IS NULL THEN Split2

ELSE Split3 END AS Subject1,

CASE WHEN Split2 IS NULL AND Split3 IS NULL THEN NULL

WHEN Split3 IS NULL THEN Split1

ELSE Split2 END AS Subject2,

CASE WHEN Split3 IS NULL THEN NULL

ELSE Split1 END AS Subject3 FROM #temp2

DROP TABLE #SplitTable

Ugh. This probably wasn't the best way to do this, and it feels kinda inelegant. On a million-row test harness, it works out acceptably in execution speed, but the splitter table would most certainly be more elegant and a lot more efficient. This reminded me of a problem I had to solve myself at some point, but the finer details were a bit more convoluted than I'd hoped.