SQL Server 2008 spatial data - fastest way to find if a given lat/long point inside an area & help with polygons

Hi. We’re playing with spatial data in SQL Server 2008 with a view to building an interface.

Basically we have a number of records in SQL Server. Each record has 2 lat long coordinates to denote a 500mx500m grid. This overlays the UK. This data has been provided to us and we don’t need to verify it.
Ie each records had min_lat, max_lat, min_long, max_lond, some_data columns

We want to pass a lat long point to sql server (ie the user clicks on a google map) and find a row in sql server where the point is inside that rows grid and  then display the relevant data for that row.

For example the following 4 lat/longs will build a grid around ben nevis:
TL - 56.798937031503 -5.00620317132339
TR - 56.798937031503 -4.99838730303897
BR- 56.7942545768597 -4.99838730303897
BL - 56.7942545768597 -5.00620317132339
 
And if we select a point – say 56.796268 -5.0034890, we want to know that it is inside the grid (which it is). Easy enough.
Is it actually faster to just do a floating point comparison on the lat/long of the passed point – ie passed lat>= min lat and passed lat <=max lat and passed long >=min long and passed long <= max long

Or should we use the SQL Server spatial geography type, build a polygon for each row and see if the point is inside the area of the row?
We where initially given the centre point of each grid and where using a proximity search to return the closest centre point to where you clicked. This worked really quickly – having fabricated a few duplicate entried to swell the data to over 3 million records, by using this code you could get the records within 1000m of the point passed in under 1 second

declare @point geography
set @point = geography::Point(53.2108154296875,-3.05141019821167, 4326)

SELECT  postcode, latlong.STDistance(@point)
FROM postcodes
where latlong.STDistance(@point)<=1000
order by latlong.STDistance(@point)

we where looking at loading a polygon for each row of data (this data is NEVER going to change, so it’s not a problem to store the polygon). Is this correct for the above lat/longs – ie a polygone with 5 points (the 5th is the start position) adding the points as below in lat long

declare @area as geography
set @area = geography::STGeomFromText('POLYGON((
56.798937031503 -5.00620317132339,
56.798937031503  -4.99838730303897,
56.7942545768597 -4.99838730303897,
56.7942545768597 -5.00620317132339,
56.798937031503 -5.00620317132339
))', 4326)

This happily accepts the polygon, but if you try to see if the point 56.796268 -4.99920317132339 is inside the polygon it returns 0 with the below

declare @point geography
set @point =geography::Point(56.796268, -5.0034890,4326)


declare @area as geography
set @area = geography::STGeomFromText('POLYGON((
56.798937031503 -5.00620317132339,
56.798937031503  -4.99838730303897,
56.7942545768597 -4.99838730303897,
56.7942545768597 -5.00620317132339,
56.798937031503 -5.00620317132339
))', 4326)

select @point.STIntersects(@area)
but if you create the point with long, then lat (ie the wrong way around) it gives 1 to show it’s inside the polygon

declare @point geography
set @point =geography::Point(-5.0034890, 56.796268, 4326)


declare @area as geography
set @area = geography::STGeomFromText('POLYGON((
56.798937031503 -5.00620317132339,
56.798937031503  -4.99838730303897,
56.7942545768597 -4.99838730303897,
56.7942545768597 -5.00620317132339,
56.798937031503 -5.00620317132339
))', 4326)

select @point.STIntersects(@area)

so, I assumed that the polygone needed to have th e lat/long values reversed as I knew the point was created correctly in lat long order. but if you try to create a polygon with

declare @area as geography
set @area = geography::STGeomFromText('POLYGON((
-5.00620317132339 56.798937031503,
-4.99838730303897 56.798937031503,
-4.99838730303897 56.7942545768597,
-5.00620317132339 56.7942545768597,
-5.00620317132339 56.798937031503
))', 4326)

It says this is not a bvalid geography instance.

So what am I doing wrong here.

Even with a spacial index, would trying to identify if a point is within a polygon in any of 3million plus records be any faster than just checking the float values against each other fot the lat and long values?