There have been several interesting replies on this often argued topic, and Gordon Uber's web site and link thereon provides a lot of material - thanks.
My own observation is the feeling that old Latin church accounts I have looked at (not all that many) do seem to have used iiii or iiij rather than iv and it was perhaps just the more common usage in mediaeval England - as well as being better visually on the dial. Having said that, one can find a lot of interesting "rogue" numbers on sundials. For example, I think the one on the church tower at Warmwell, Dorset, dated 1646, has VIIII incised for 9 and I have seen several old vertical dials where it is not that the lower numerals have just been inverted (these are sometimes known as "tumbling hours" on clock dials and are quite common with Arabic numerals) but that the V has been inverted relative to the position of the I so that for example the 7, VII, is actually ^II, or IIV if you look at it the other way in order to read the V the correct way up. I think St Alkmund's, Whitchurch, Shropshire has a (now very faded) painted dial with several such reversed numerals and this cannot be earlier than c 1720 when it was rebuilt. The link on Gordon's web page led me to an example on a clock dial in New Hampshire having V IV IIV IIIV IX which is shown at http://members.aol.com/lolathrop/roman/iv.html . I would be interested in other comments on these "erroneous" numbers - as pointed out at the link above the Romans and others might have recognised the "inverted v", "^", easily enough as meaning v or 5, and that is fine if you count i ii iii iiii ^ ^i instead of i ii iii iiii v vi, but if the position of the i beside the v makes a difference, as it does when you use subtractive as well as additive combinations, there is scope for confusion. Andrew James N 51 04 23 W 1 17 46