When converting disk images from other formats to vmdk using qemu-img, the vmdk disk produced is always a monolithic sparse disk with an adapter type of "ide" and disk geometry to match. Depending on the partitioning scheme of the guest OS, such a disk, even after being converted to a format compatible for ESX use, will still often be unbootable if attached to the virtual SCSI controller instead of the IDE controller.
This behavior currently leads to a hard requirement that the vmware_adaptertype=ide image property be set when the vmdk is uploaded to glance. Failure to set this property, a step often overlooked by the user, will often lead to guest boot failure that is hard to track down. In some cases, vmdk disks uploaded to glance even lacks the DDB metadata that indicates the adaptertype of the disk, which complicates serviceability further.
I am looking into alternative means to update the disk geometry of the disk appropriately for use by either controller.