The Head-to-Head Graphs
Below you'll find two graphs: one graph of the polling average for Royal vs. Sarkozy from January 2 to May 4, and one graph featuring the polling average and the actual polls for the last two weeks of the election.

If the graphs were right, the election would have ended somewhere in the ballpark of:
Royal: 46%
Sarkozy: 54%

The actual election result was:
Royal: 46.9%
Sarkozy: 53.1%

In other words, the polling average was very close, if slightly underestimating Royal. The poll having Royal and Sarkozy almost on parity was conducted by Tunisian pollster 3C, newcomers to the scene. Better luck next time guys!
As can be seen from the graphs, Sarkozy lead the head-to-head for much of the duration of the campaign. Right before the first round, the gap between Royal and Sarkozy decreased significantly, only to immediately return to it's previous levels in the aftermath of the first round.
The polls were then tightening marginally up until the debate. Until that point, there were still a select few of us who remained optimistic that Royal could pull this off, but the poll numbers following the debate, which were giving Sarkozy a victory margin in landslide territory, forced even the most hopeless of optimists among us (like yours truly) to concede that the election was pretty much over.

The last few polls before the election were as follows:

	BVA	CSA	IFOP	IPSOS	LH2	TNS-S.
Date	4.5	3.5	4.5	4.5	30.4	4.5
Royal	45.0%	47.0%	47.0%	45.0%	48.0%	45.5%
Sarkozy	55.0%	53.0%	53.0%	55.0%	52.0%	54.5%

The pollsters who got the closest were thus CSA and IFOP, though all of the pollsters were reasonably close.


Crossover Votes
In the last two weeks, the pollsters also regularly polled how Bayrou and Le Pen voters would vote in the second round, following the elimination of their respective candidates in the first round. The two graphs below show how the crossover votes changed during the last two weeks (averages and actual poll numbers) for Bayrou and Le Pen voters, respectively:


One caveat: Some pollsters didn't report how many of the voters would abstain, but merely the percentages for the candidates of the voters intending to vote in the second round (thus, the sum of the percentages for the candidates would equal 100%, and the people abstaining were simply discarded). For a meaningful comparison, I thus had to commit the cardinal sin of recalculating the numbers accordingly for the pollsters that did report the number of absentions. In both graphs I've also plotted the number of absentions where reported. It's obviously not directly comparable to percentages for the candidates, but it gives us some idea of how the sympathies changed during the last two weeks.

According to the graphs, Bayrou voters would break about evenly for Royal and Sarkozy, whereas Le Pen voters would behave as expected by voting overwhelmingly for Sarkozy.
Exit polling indicated the following break-down among previous Le Pen and Bayrou voters:

Bayrou voters
Royal: 41% (49%)
Sarkozy: 42% (51%)
Abstentions/Blank: 17%

Le Pen voters
Royal: 15% (19%)
Sarkozy: 63% (81%)
Abstentions/Blank: 23%

Numbers in parenthesis excludes abstentions, for comparison's sake. (Average of the following sources: IPSOS (pdf), TNS-SOFRES, CSA (pdf))


What If...?
Prior to the first round, a few of the pollsters also polled other potential second rounds; most frequently polled was Sarkozy vs. Bayrou, and a distant second was Royal vs. Bayrou. Below, the graphs for the respective match-up:


Do note that the second graph is not of polling averages, but of the actual polls, as there were too few polls released to plot any sort of meaningful average. The averages in the first graph consists primarily of IPSOS polls, as they were the only ones polling Bayrou vs. Sarkozy on a regular basis.


Conclusions
Over all, the polls very close to the actual results. The one big mistake all the pollsters did make this time around was severely overestimating Le Pen's poll numbers in the first round. In 2002, they underestimated him; in 2007, they overestimated him. In the second round, it appears they were all in the right neighbourhood.


Oh, One More Thing...
The first two polls I have in my Excel file were done on October 13, 2006, by IFOP and TNS-Sofres. Both showed Royal losing to Sarkozy 47% - 53%.