I'm sure you'll realize at some point that this is subjective and anecdotal, but caveated (I made up that word) it is based on twenty-some years of experience with deploying crap to thousands of computers.
Good: Usually works
Bad: Sucks when you have to install 500 or 1,000 within 24 hours
Good: Pretty Simple
Bad: Only works with MSI installers, MSP patches, and ZAP wrapped packages. No phased deployments. No status monitoring capability. No bandwidth throttling. No automated distribution management (regional/local server shares)
Good: Customizable, Flexible.
Bad: Requires some knowledge and experience. More work required to implement monitoring, phasing, and distribution automation management. Difficult to manage remote/mobile clients
Good: Usually works. Customizable, Flexible. Status monitoring. Phased deployments. Automated distribution management. Manages remote/mobile clients and provided automated bandwidth throttling.
Bad: Additional cost. Additional learning curve.
Snapshot Imaging (Ghost, etc.)
Good: Provides a consistent result for all newly-imaged computers
Bad: Breaks quite a few applications due to replicated information. Increases imaging process time. Increases base image build and update time.
Sequential Imaging (SCCM task sequence + OSD)
Good: Provides a consistent result for all newly-imaged computers. Usually eliminates problems incurred with duplicate identification issues commonly encountered with snapshot imaging techniques. Can often employ existing SCCM packages without the need to repackage twice.
Bad: Requires some learning. Increases imaging process time.
Application Virtualization (App-V, ThinApp, etc.)
Good: Provides encapsulation and portability. Provides isolation and prevents resource and version conflicts
Bad: Often costly to implement. Additional learning curve. Not all products work well with it (most do).
Server Hosted / Terminal Emulation (Remote Desktop, Remote App, Citrix, etc.)
Good: Shared from a single installation. One target to maintain and upgrade.
Bad: Costly to implement. Single point of failure*. Some vendors prohibit this option in their license terms. (* even with clustering and load balancing, this is similar to the risks of cloud services)