One angle I feel is often missed in the whole cancer discussion is that it's a multidimensional problem.
1) Developing drugs which can damage cancer cells selectively.
Like selecting a few needles in a haystack, cancer cells typically "look" the same as healthy cells, which makes selectively destroying them very difficult. Some cancers (such as CML) have a specific driver mutant proteins which allows for a pinpoint attack (Gleevec), but if this target protein mutates and Gleevec can no longer bind then that drug instantly becomes totally ineffective.
2) Identifying malignant tissue
Surgery is still one of the most important tools against many forms of cancer, yet being able to identify malignant tissue vs. healthy stuff has always been a problem. In the first half of the 20th century the radical mastectomy was all the rage, but in reality provided little benefit. I was lucky enough to see Roger Tsein speak a few years ago (Nobel prize for GFP) who is pioneering a way to fluorescent tag malignant tissue which can be viewed in real time to give surgeons an augmented reality overview of a tumour to maximize the chance of getting all the malignant cells.[http://en.wikipedia.org/wiki/Roger_Y._Tsien#Fluorescence-ass...].
3) Stopping metastasis
Even if a patient presents with a tumour and that tumour is removed, it's impossible to tell if any of the maligant cells have managed to escape to other parts of the body, where they slowly start to regroup before launching a subsequent attack. This is the primary reason why cancer survivors have 5Y and 10Y survival rates as opposed to, "You're cured". The mechanism and time in a cancer at which this happens depends on so many factors its currently almost impossible the predict.
4) Drug delivery
The tumour micro environment is so foreign compared to the normal stromal environment, and moreover so heterogeneous between tumour types (which in turn depends both on a cancer's underlying genotype and its associated tissue, vascularization and a wide range of additional factors) that creating drugs which can just survive long enough to act on their target can be difficult. This, combined with the fact that tumours are, compared to normal tissue, often poorly vascularized, means just getting drugs in can be a major challenge. I remember reading how often the vascularization of a tumour can be proportional to rate of growth and inversely proportional to chemotherapeutic efficacy (although don't quote me because I can't find the reference) meaning smaller, slower growing tumours often represent those most difficult to treat while larger, more aggressive ones may respond better, if caught in time.
5) Cancer is effectively microscale evolution
Perhaps the biggest problem is that once cancer cells begin to acquire some initial mutations (and lie in a pre-cancerous form) they're often more susceptible to further mutations. In a way, this allows them to employ a sort of bet-hedging strategy, such as that seen by yeast or other single celled organisms. No longer can we treat the cancer cells as part of our multicellular body, but as a separate, single celled population. This means that even if certain drugs are effective, there will be a small population of cancer cells who may have mutated in such a way that they are resistant, so even if a treatment gets 99% of the cells, that final 1% can restart and the same bet-hedging strategy is re-employed to create another, diverse set of cells. This is totally analogous to antibiotic resistance. The range of this diversity varies significantly between cancers, but as we "pick out" the easier ones with an obvious target, this will become an every increasing issue.
These are just a few points - there are more, but I've tried to focus on ones not yet brought up in discussion. This is an area I have some experience with, and if anything is unclear let me know and I'll do my best to explain.
[EDIT]: This is not meant to be quite as, "we're all doomed" as it appeared. I think Dn_Ab's post summarizes how I feel more accurately. The development in a range of different areas has been staggering (childhood leakemia, CML, Her2+ BC etc), and the reason this is such a daunting task is because the magnitude has only reared its large, complex head in the last 10-15 years or so. Despite the challenges, it is a very exciting time to be a cancer biologist.